The knockout plasmid was constructed using polymerase chain reaction (PCR) and seamless cloning technology. The primers used in this study are presented in Table 1. First, the vector skeleton was generated using an ApaI-XbaI (Vazyme Biotech Co., Ltd.)-digested pLL plasmid. Three fragments, 1,300 bp homologous arms upstream and downstream (obtained by PCR using primers up/down-F/R with L. paracasei CGMCC4691 genome as a template), and sgRNA (obtained by PCR using primers sgRNA-F/R with pLL as a template) were generated and purified using a gel purification kit (Axygen, SV, USA), all of which were connected to generate up–down-sgRNA fragments. The resulting fragment was ligated into the vector skeleton to produce pLL-07315 using a seamless cloning kit (Vazyme Biotech Co., Ltd.) and verified using ldhYZ-F/R following transformation into E. coli Top10 competent cells.

A competent cell of L. paracasei CGMCC4691 was prepared as described by Song et al. (2017). The constructed plasmid was electroporated into cells using a GenePulser Xcell (Bio-Rad) and a 2 mm cuvette (BTX) at 2.0 KV, 200Ω, and 25 μF. After recovery for 3 h in the MRS medium, the cells were separated on agar plates and grown for 72 h. A single strain was selected and verified by PCR and named pLL: 07315. Precise sequencing was also performed using BGI.

Skim milk (10% w/v) was prepared from skim milk powder (Shijiazhuang Junlebao Dairy Co., Ltd.). Skim milk powder was dissolved in 50°C sterile water and autoclaved at 115 °C for 15 min. After cooling to 37 °C, 400 ml of milk was transferred into six different sterile flasks. Pre-cultured L. paracasei CGMCC4691 and its mutants were inoculated in triplicate at a 5% inoculum size (2×10⁷ CFU). The mixed milk was fermented until the pH reached an end-point of 4.5. Subsequently, a sample from each flask was collected before (4 h) and after (8 h) the fermentation periods and end-points. Before analysis, these samples were stored at −80 °C.

pH was measured using a pH meter (METTLER TOLEDO, China). TA was determined using 0.1 mol/L NaOH with phenolphthalein as a color indicator, and the result was expressed as °T.

The viable counts of different periods during milk fermentation were determined as follows: 1 ml of the sample was diluted in 9 ml of normal saline (0.85%) by vortex oscillation, and the dilution was performed continuously. The diluted liquid (100 μl) was cultured using the MRS agar plate pouring method. The viable counts were calculated after incubation at 37 °C for 48 h and expressed as CFU/ml.

The fermented milk samples stored at −80 °C were thawed. We transferred 50 mg of each sample into a fresh centrifuge tube, and 400 μl methanol solution (containing 5 μg/ml L-2-chloro-phenylalanine as the internal standard) was added and vortexed for 1 min. The sample was then mixed two times at 60 Hz for 3 min and centrifuged at 13,000 rpm at 4°C for 10 min using a cryogenic centrifuge (Thermofisher FRESCO21, USA). Supernatants were collected for analysis. Equal volumes of all samples were mixed as quality control samples. All reagents were analytically pure and purchased from Shanghai.

The samples were analyzed on an Agilent 1290 infinity (USA), incorporating a HSS T3 chromatographic column (water, 2.5 μm, 100^*2.1 mm). The mobile phases A and B consisted of water and acetonitrile (mixed with 0.1% formic acid). The flow rate of the mobile phase was 0.4 ml/min at a column temperature of 40°C. The injection volume was 4 μl. The optimized gradient elution was 0–3 min, 20% B; 3–9 min, 20–95% B; 9–13 min, 95% B; 13–13.1 min, 95–5% B; and 13.1–16 min, 5% B.

Mass spectrometry (MS) analysis was performed using a Q-TOF-MS^E (Agilent 6545 UHD, USA), coupled with positive (ESI+) and negative (ESI–) modes. The capillary voltage was 4.5 kV and 3.5 kV in ESI+ and ESI–, respectively, and the drying gas flow was 8 L/min and 10 L/min at a temperature of 325°C, respectively. The nebulizer voltage was set at 20 psi. The fragmentation voltage was set to 120 V. The MS scans were performed over an entire information range of 50–1,500 m/z.

OPLS-DA, a common classification method, was performed to identify the differences between the two groups, which exploits dimension reduction by combining regression models; the regression results were analyzed using a discriminant threshold. A total of 1,096 and 927 substances were generated from the fermented milk in the ESI+ and ESI– modes, respectively. The OPLS-DA of fermented milk is shown in Figure 3. The abscissa represents the principal component score during the quadrature signal correlation; the farther the location of the two milk samples, the greater the difference between the two groups. The ordinate represents the quadrature component score value; the farther the parallel samples are, the greater the difference within the group. The milk fermented by the WT strain at 4 h (W1) was located away from the milk fermented by the mutant strain (M1), indicating that the two groups had significant differences, and a total of 52.4% was explained by the dataset. The critical parameters used to evaluate the OPLS-DA model are presented in Table 2. In the ESI+ model, the R²X, R²Y, and Q² values were 0.524, 1, and 0.654, respectively. In the ESI– model, the R²X, R²Y, and Q² values were 0.632, 0.999, and 0.762, respectively, where R² and Q² represent the explanatory ratio and predictability, respectively. The closer the value is to 1, the higher the model accuracy. Generally, when R² and Q² >0.5, the model fits well.

To select the differential metabolites, a volcano plot was generated from the OPLA-DA filtering results (Figure 3B). In ESI+, the red dots indicate that the metabolites were upregulated, with FC >1.1 and P ≤ 0.05, and the blue dots indicate that the metabolites were downregulated, with FC <0.9 and P ≤ 0.05. The farther the dot is from the origin, the greater the difference between the two groups. A total of 51 metabolites were identified. Next, these potentially differential compounds were further used for enrichment analysis by (1) KEGG ID conversion using the MBRole online database and (2) importing the ID into the 4,691 KEGG metabolic pathways for pathway enrichment analysis. This resulted in 31 compounds, and a heatmap was generated to show their relative abundances (Figure 3C). Enrichment analysis of the two groups was performed using L. paracasei CGMCC4691 (lpq), and nine differential metabolites were enriched in the pathways. The results are shown in Figure 3D, and the metabolites are presented in Table 3. The key pathways include phenylalanine, tyrosine, and tryptophan biosynthesis, as well as phenylalanine metabolism, pyruvate metabolism, and nicotinate and nicotinamide metabolism.

A method similar to that described above was used to identify differential metabolites between the two sample groups (M2 and W2) at 8 h. In total, 1,117 differential metabolites were identified as potential candidates (VIP >1) in the OPLS-DA plot, and significant separation was observed in the positive mode (Figure 4A). The model evaluation parameters (R²X, R²Y, and Q² were 0.806, 1, and 0.896 for ESI+; and 0.541, 0.994, and 0.837 for ESI–) indicated that the model was both reliable and valid. In ESI+, the volcano plot showing the FC of these candidates is presented in Figure 4B, and the criteria were FC >1.1 or ≤0.9 and P ≤ 0.05. We obtained 207 differential substances from the positive model. The dot represents a variable, and the farther the variable is from the origin, the greater the contribution to the difference between the two groups. Subsequently, the KEGG ID of 80 variables was obtained from the MBRole database, and a heatmap was generated to visualize these substances (Figure 4C). Pathway enrichment analysis was performed using these metabolites (Table 4). A total of 18 signaling pathways are shown in Figure 4D, and the top signaling pathways were phenylalanine metabolism, cysteine and methionine metabolism, biotin metabolism, aminoacyl-tRNA biosynthesis, and pyruvate metabolism.

The differential metabolites in fermented milk at pH 4.5 were compared using the multiple analyses described above (Figure 5). A total of 1,132 were identified, and OPLS-DA showed significant separation in the positive mode (Figure 5A). The model evaluation parameters (R²X, R²Y, and Q² = 0.592, 0.997, and 0.759, respectively, in ESI+; and 0.584, 0.996, and 0.73, respectively, in ESI–) are presented in Table 2, which indicates that the model has good stability and predictability. In ESI+, the volcano plot showed 123 differential metabolites according to the criteria mentioned above, where the red and blue dots represent significantly differential metabolites (Figure 5B). Next, two statistical analysis methods were performed to further select featured substances: 49 KEGG IDs were obtained from the MBRole database, and a heatmap was used to visualize their relative abundance (Figure 5C). Pathway enrichment analysis was performed using these metabolites, and the details are presented in Table 5. A total of 20 signaling pathways are presented in Figure 5D, and phenylalanine, tyrosine, and tryptophan biosynthesis, as well as biotin metabolism and pyruvate metabolism were involved in key metabolic pathways.

In total, nine differential metabolites were detected in the mutant strains during pre-fermentation when compared to the WT strain. Figure 3D shows the results of the enrichment analysis. The metabolites involved in pyruvate, amino acid, and pyrimidine metabolism included (S)-lactate, phenylpyruvate, L-proline, 5′-methylthioadenosine, and indole-3-acetaldehyde, all of which were downregulated. (S)-lactate, which is catalyzed by lactate dehydrogenase in the last step of glycolysis, provides energy to cells and is the main substance that adds flavor to fermented milk. The ΔAF91_07315 strain failed to reach the exponential growth phase during pre-fermentation, which may be related to reduced lactate metabolism, insufficient energy supply, and delayed growth of the strain (Rico et al., 2008; Jingjing et al., 2021). In addition, proline, phenylpyruvate, and indole-3-acetaldehyde were downregulated, and the growth-promoting effect of proline on Lactobacillus has been demonstrated (Wang et al., 2021a). Sun et al. (2019) added amino acids, uracil, and other substances during the fermentation of L. rhamnosus, which improved the start-up speed. In this study, phenylpyruvic acid, indole-3-acetaldehyde, and 5′-methylthioadenosine as the precursors of phenylalanine, tryptophan, and methionine were less synthesized, which affected the growth of the strain (Shapiro, 1962; Yonezawa et al., 2010; Meng et al., 2021).

During mid-fermentation, the largest gap in pH and viability was observed when compared to the WT strain, and the metabolites detected at this time were mainly L-methionine, 5′-methylthioadenosine, L-cysteine, phenylpyruvate, phenylacetic acid, L-lysine, and L-lactic acid. In particular, the gap in lactic acid content increased compared to the previous stage, which indicates the major role of lactic acid production in regulating the pH of fermented milk (Morelli et al., 1986; Gaspar et al., 2013), and the knockout of ldh had a significant effect on lactic acid production. Nevertheless, lactic acid was still detected throughout the fermentation phase in the mutant strain, probably owing to the synergistic effects of other genes encoding acid-producing enzymes (Viana et al., 2005). The synthesis of nutrients, such as amino acids, in LAB is limited to fermented milk; therefore, exogenous nitrogen sources are used to promote their growth, such as the breakdown of amino acids and peptides. In this study, the L-cysteine content was reduced in the ΔAF91_07315 strain except for methionine, phenylalanine, and its precursors. Wang et al. (2021b) reported different pathways of amino acid biosynthesis at different stages of fermentation, and the synthesis of L-cysteine may be due to different protein hydrolase activities.

Fermentation was stopped at pH 4.5 to identify the various differential metabolite strains and the differential metabolites detected were all upregulated in the ΔAF91_07315 strain, including malonyl-CoA, L-lysine, phenylpyruvate, riboflavin, alpha-isopropylmalate, and spermidine, which may be due to the prolonged fermentation time of the mutant strain. In addition, alpha-isopropylmalate and spermidine were involved in pyruvate and arginine metabolism, respectively, providing energy and improving the nutritional quality of fermented milk. Malonyl-CoA is involved in fatty acid synthesis and carbon metabolism, which is related to yogurt flavor (Foster, 2012). Riboflavin is a trace organic substance required for growth and metabolism that stimulates the growth of LAB (Sun et al., 2022). The fermented milk of the ΔAF91_07315 strain did not complete fermentation when the WT strain was fermented at a pH of 4.5, after which the viable counts of the ΔAF91_07315 strain continued fermentation and reached the same pH as the WT strain, probably because riboflavin, L-lysine, alpha-isopropylmalate, and spermidine compensated for the growth defect in the ΔAF91_07315 strain. Further investigation is required because the metabolites of the mutants were upregulated at the end of fermentation, some of which may be toxic, and it is unclear whether there are biosafety issues with fermented milk.

In this study, the ldh gene of L. paracasei CGMCC4691 was knocked out using CRISPR gene editing technology. The fermented milk results showed that the ΔAF91_07315 strain had slower growth and acidification capacities than the WT strain. At the initial stage of fermentation, the WT strain entered the exponential growth phase, and metabolites, including (S)-lactate, phenylpyruvate, L-proline, and indole-3-acetaldehyde, were downregulated compared to the WT strain, which affected the start-up speed of the ΔAF91_07315 strain. At the mid-stage of fermentation, pH and viable counts were the most disparate, with (S)-lactate being the main cause of the differences in pH. In addition, L-cysteine and precursor metabolites involved in amino acid metabolism were reduced because of the influence of protease activity. Although the fermentation time of the mutant strain was prolonged at the end of fermentation, malonyl-CoA, L-lysine, the amino acid intermediate metabolites, phenylpyruvate, and spermidine were upregulated, contributing to the formation of flavored substances. Upregulation of metabolites, such as riboflavin, compensated for the growth defects of the mutant strain. In conclusion, lactate dehydrogenase plays an important role in the growth and acidification of L. paracasei and affects the production of flavored milk substances.